ClawHub
Back to skill

Security audit

Planning Validator

Security checks across malware telemetry and agentic risk

Overview

This skill appears to be a harmless stub, but it presents itself as a safety validator that can approve plans without actually validating them.

Install only if you treat this as a prototype or checklist, not as a real execution gate. Do not rely on it to approve important or risky agent plans until it actually reads plan files, validates tools and permissions, checks dependencies and constraints, and fails closed when validation is incomplete.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Description-Behavior Mismatch

Medium
Confidence
95% confidence
Finding
The skill advertises pre-execution validation of tool availability, permissions, dependencies, and feasibility, but the implementation never reads or validates the provided plan file and performs only minimal structural checks. In a security-gating role, this creates a dangerous false sense of assurance: downstream agents may proceed with unsafe or impossible plans because this validator reports success without enforcing the claimed controls.

Intent-Code Divergence

Medium
Confidence
98% confidence
Finding
The check_reality function always returns a hardcoded successful result with high confidence, despite claiming to assess whether a plan is realistic. Because this skill is meant to validate plans before execution, the unconditional success response can allow hallucinated, non-executable, or policy-violating plans to be treated as trustworthy and executed without meaningful review.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.